基于BlazePose和随机森林算法的异常步态检测  

Abnormal Gait Detection Based on BlazePose and Random Forest

作　　者：黄灶荣 王春宝 韦建军[1] 余学书 谭啸海 HUANG Zaorong;WANG Chunbao;WEI Jianjun;YU Xueshu;TAN Xiaohai(Guangxi University of Science and Technology,Liuzhou,Guangxi 545616 China;Guangdong Mingkai Medical Robot Co.,Ltd.,Zhuhai,Guangdong 519075,China)

机构地区：[1]广西科技大学,广西柳州545616 [2]广东铭凯医疗机器人有限公司,广东珠海519075

出　　处：《计算技术与自动化》2024年第2期62-69,共8页Computing Technology and Automation

基　　金：国家自然科学基金资助项目(61963007)。

摘　　要：异常步态对行动能力产生严重影响,因此,及时、自动地检测异常步态具有至关重要的意义。本文提出了一种基于BlazePose和随机森林算法的人体异常步态检测方法。先利用BlazePose算法提取RGB视频中的人体骨骼关键点,然后通过数据处理获取7个关键的步态特征参数。最后采用随机森林算法作为步态分类器,用于区分正常步态与异常步态。利用142例异常步态数据和257例正常步态数据对分类器进行训练和测试评估,实验结果显示准确率和召回率分别达到97.5%和90%,表明该方法在异常步态检测方面具备一定的可行性和实用价值。Abnormal gait significantly affects mobility,making timely and automatic detection of abnormal gait critically important.This study proposes a human abnormal gait detection method based on BlazePose and the random forest algorithm.Firstly,the BlazePose algorithm is utilized to extract skeletal keypoints of the human body from RGB videos.Next,seven key gait feature parameters are obtained through data processing.Finally,the random forest algorithm is employed as the gait classifier to differentiate between normal and abnormal gaits.The classifier is trained and evaluated using 142 cases of abnormal gait data and 257 cases of normal gait data.Experimental results show an accuracy of 97.5%and a recall rate of 90%,indicating that the proposed method exhibits certain feasibility and practical value in the field of abnormal gait detection.

关 键 词：BlazePose 随机森林 异常步态检测 数据处理 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]